FreeRAG is a tiny native macOS menu bar app that turns screenshots, clipboard fragments, and voice notes into a local corpus for coding agents and LLM workflows.
It is built for people who use Codex / Claude Code and keep losing useful context in screenshots, browser tabs, copied text, meeting notes, and half-finished thoughts.
Download DMG · Quickstart · Demo · FAQ · Latest Release · Launch Kit
- Collect raw context from the menu bar HUD: screen, clipboard, or voice.
- FreeRAG writes everything locally under
~/Documents/Corpus/. - Ask MyRAG in Codex / Claude Code to mine recent material.
- MyRAG folds duplicates, reads representative evidence, and returns a table: one matter per row.
- You decide what should become a worklog, project note, or cleaned-up raw entry.
- Download DMG
- Latest GitHub Release
- Current beta:
0.5.1, build3 - macOS app: Swift / AppKit, menu bar only
This beta build uses a self-signed local stable signing identity and is not Apple Developer ID signed or notarized. macOS may show Gatekeeper warnings.
FreeRAG collects. MyRAG reads and mines.
FreeRAG does not upload screenshots, clipboard content, or voice recordings by itself. The local corpus is intended for explicit use by the user through Codex / Claude Code and the bundled MyRAG skill.
MyRAG output is table-first: one matter per row, with facts, numbers, people/projects, judgment, risk, next step, evidence, and confidence merged into the same row. Large repeated clipboard images are folded by exact SHA-256 before reading representative samples.
- Captures local raw material into a consistent corpus folder.
- Keeps the macOS app small: no automatic cloud analysis, no default OCR, no default transcription.
- Lets MyRAG read the local corpus and produce a matter-by-matter summary table for the user.
- Marks raw material as done only after the user has reviewed the summary and confirmed that the raw evidence has been taken over.
FreeRAG/
Packaging/ Build scripts for the native app and DMG.
Resources/Assets/ App icon and status assets.
Sources/main.swift Swift/AppKit app implementation.
shared/skills/myrag/ MyRAG skill and local corpus helper script.
docs/ Product docs, public status, and overview pages.
release/github/ GitHub publication plan, checklist, release notes.
FreeRAG/Packaging/build_native_app.sh
FreeRAG/Packaging/build_dmg.shLocal build output goes to dist/. dist/ is intentionally ignored by Git; public binaries should be attached to GitHub Releases instead of committed to the source repository.
Current app version: 0.5.1, build 3.
- Download the DMG from a GitHub Release.
- Drag
FreeRAG.appto/Applications. - Launch it from
/Applicationsso macOS permissions stay tied to a stable app location. - Grant Screen Recording, Accessibility, and Microphone permissions when needed.
- Copy the bundled
myragskill into your Codex / Claude Code skill directory if you want LLM-side corpus mining.
~/Documents/Corpus/
_index.json
_library.json
README_FOR_LLM.md
screen/
clipboard/
voice/
processed/
processed/ is optional and should contain only high-density summaries, worklogs, project notes, or confirmed representative evidence. MyRAG should not convert hundreds of noisy raw entries into hundreds of processed directories.
See PRIVACY.md. Short version: raw evidence stays local by default. Do not commit or publish a corpus folder, screenshots, recordings, transcripts, API keys, or private team/customer data.
See SECURITY.md for reporting and release-safety notes.
MIT. See LICENSE.